← sources

ChatGPT fabricated a dishwasher adjustment mechanism, generated a false instructional diagram for it, and gave me physical instructions while I was underneath the appliance

source: reddit (reddit:ChatGPT)published 2026-09-05 16:24ingested 2026-09-06 02:36

retained text

I am posting this because I think this is an important physical-safety failure, rather than simply another funny example of an AI hallucination. I was trying to remove my integrated Teka DFI 46700 dishwasher from underneath my kitchen countertop so that I could inspect it. I did not know how the dishwasher was secured or how its levelling feet worked, so I asked ChatGPT for help and supplied photographs throughout the process. What followed concerned me enough that I think other ChatGPT users should know about it.

  
**1. It started with a completely ordinary request** 

First I photographed the plumbing underneath my sink because I needed to shut off the dishwasher’s water supply. I then photographed the underside of the dishwasher and asked how to lower it so that I could pull it out. At this stage I was physically working on the appliance while following ChatGPT’s instructions. 

**2. ChatGPT identified components and told me what to manipulate** *(Figure 10)*

I supplied close-up photographs of the left and right underside. ChatGPT identified the visible white components as front levelling feet and told me to retract them. It also made and later revised other component identifications. The important issue was what happened when it still did not know the answer. 

**3. I gave ChatGPT the exact model: Teka DFI 46700**

ChatGPT found a Teka specification stating that the dishwasher has “adjustable back feet from front.” From that point it repeatedly treated the specification as proof that there had to be a front-accessible rear-foot mechanism that we simply had not found. But it had not obtained the model-specific installation/service diagram showing where that mechanism was.

**4. I kept supplying photographs because the dishwasher would not move** *(Figure 1)*

I removed the screws securing the dishwasher to the countertop and completely retracted both visible front feet. The dishwasher still would not move or rock. This should have been the stop point: the appliance was contradicting the model’s prediction and the actual adjustment mechanism remained unverified.

**5. Then ChatGPT generated a mechanical instruction diagram** *(Figure 2)*

Because I could not find the installation sheet, I asked ChatGPT to illustrate what it meant. It generated a polished instruction graphic showing a centre-front “Rear foot adjuster,” a screwdriver inserted into that supposed component, and specific clockwise/counter-clockwise directions. ChatGPT had never established that this component existed there. That is the part I find most serious. The diagram was not copied from a Teka service manual. It was a visual representation of ChatGPT’s own unverified hypothesis, presented in the visual language of technical documentation.

**6. I eventually found the real rear foot myself**

After spending a considerable amount of time on the floor trying to understand why the machine would not move, I managed to photograph a real levelling foot at the rear. That does not by itself prove that Teka’s “front adjustment” feature is wrong - there may be a linkage - but ChatGPT had never found or verified that linkage. It had simply drawn a specific centre control as though it were real. ChatGPT then admitted that the generated diagram “showed a centre front adjuster that I had not actually verified on your machine” and told me not to use the image as a mechanical reference. 

**7. Why I think this matters beyond my dishwasher** 

The failure was not just “AI gave a wrong answer.” The sequence was: uncertain visual identification → inference from an incomplete specification → repeated confident statements → physical manipulation instructions → failure to stop when reality contradicted the prediction → an AI-generated technical illustration of an unverified mechanism → specific tool-placement and rotation instructions. I was physically working around the appliance in real time. A built-in dishwasher contains heavy moving mass, sheet-metal edges, electrical parts and plumbing. Another user might reasonably trust a polished diagram and reach into a blind area to find the illustrated component.

*I am not claiming that an injury occurred. I am saying the system created a foreseeable physical-safety risk by failing to distinguish “I think this is how it works” from “put your tool here and turn it this direction.”*

[\(Figure 1\) Original Photo I uploaded -  Full underside with visible front feet retracted.](https://preview.redd.it/uvc1tuan8qnh1.png?width=2048&format=png&auto=webp&s=bec4735e6ba60760a20219a03d89d5f64795f41f)

[\(Figure 2\) ChatGPT-generated unverified instructional graphic. DO NOT USE AS REPAIR GUIDANCE. WARNING: This is preserved as evidence of the incident. The centre-front rear-footadjuster shown in this generated image was not verified.](https://preview.redd.it/1b9i1a3e8qnh1.png?width=1536&format=png&auto=webp&s=9e527fb4b36ab559443c093f796412f8646f113d)

document fetched 2026-09-06 00:19 · canonical https://reddit.com/r/ChatGPT/comments/1w857a9/chatgpt_fabricated_a_dishwasher_adjustment